We are experiencing a widespread shift toward production-scale AI infrastructures, where protection must be built right from the start into AI and SOC workflows. That is why CrowdStrike and NVIDIA are expanding their collaboration with the aim of protecting AI agents.
As part of the annual NVIDIA conference taking place in San Jose, California, both companies have reached an agreement to promote controls that take into account user intent and govern how agents plan and execute tasks, allowing flexible autonomy while limiting the impact of unwanted or malicious behavior.
The architecture integrates security natively into the AI agent stack, enabling organizations to operationalize autonomous systems with governance, visibility and control from development to execution, regardless of where the agents run.
AI agents pose new security challenges
As organizations move from using co-pilots to AI agents that think, reason and act autonomously at machine speed, security models must evolve. AI agents pose a fundamentally different security challenge, possessing privileged identities with direct access to data, applications, computing resources, and other agents. Traditional static controls were not designed to manage systems that move at the speed of AI. Protecting AI agents requires continuous enforcement of security across the entire AI system, not one-time checks that run at machine speed.
By integrating the Falcon platform directly into the NVIDIA OpenShell runtime, the security-integrated AI design enables security to be built into the foundation of autonomous systems. As part of the NVIDIA Agent Toolkit, the open source OpenShell runtime environment provides sandboxed environments with private inference and integrated policy enforcement.
The Falcon platform extends protection to on-premises agents running on NVIDIA DGX Spark or NVIDIA DGX Station, and can also extend security to cloud agents based on the NVIDIA AI-Q open source design for advanced research.
Organizations will gain unified visibility and continuous monitoring and enforcement at runtime to limit unsafe behaviors, prevent immediate tampering, and enforce policies across the entire AI lifecycle.
Key capabilities of the “Secure by Design” AI model include:
- Enforce AI policies across the entire agent stack: Falcon AI Detection and Response (AIDR) will integrate with the OpenShell runtime to protect every request, response, and agent action in real-time.
- Endpoint protection for on-premises AI agents: Falcon Endpoint Security will protect on-premises agents on NVIDIA DGX Spark or DGX Station running OpenShell, applying host-level controls and continuous behavior monitoring across system activity and agent execution.
- Cloud runtime protection for AI agents: Falcon Cloud Security will protect agents built on the NVIDIA AI-Q Blueprint in cloud and data center environments, providing unified visibility and runtime controls across AI infrastructure and workloads.
- Identity-based governance for agent access: Falcon Next-Gen Identity Security will provide dynamic identity management for on-premises agents, enforcing access controls across data, APIs, and services so agents operate within defined privilege boundaries.
Secure Architecture by Design
“As we enter the age of agents, agents are no longer just assisting, they are acting,” said Daniel Bernard, chief commercial officer at CrowdStrike. «This change radically modifies the security equation, and this must be integrated into the AI architecture itself. “Together with NVIDIA, we are delivering a secure-by-design architecture that enables organizations to deploy agents with confidence and control.”
The Falcon platform extends protection to on-premises agents running on NVIDIA DGX Spark or NVIDIA DGX Station
“Autonomous agents will radically transform the way we work,” said Justin Boitano, vice president of Enterprise Platforms at NVIDIA. “By integrating CrowdStrike’s security platform with NVIDIA’s agent toolkit, we enable enterprises to create and scale autonomous, more secure AI agents to help transform their operations, empower every employee, and securely generate intelligence at the speed of business.”
“AI infrastructure is moving from experimentation to mission-critical production,” said James Higgins, chief information security officer at CoreWeave. “As we expand GPU-accelerated environments, AI agents must be observable, controllable, and resilient by design. “The collaboration between CrowdStrike and NVIDIA protects AI systems at their foundation, enabling high-performance AI environments without compromising control.”
The Secure-by-Design AI Blueprint model reinforces CrowdStrike’s position as a cybersecurity provider for enterprise AI, integrating security directly into the AI stack, wherever it resides.
